Forecasting future values
Once we have successfully trained the XGBoost model on historical time series data, the next logical step is to use this model to predict future values. However, in time series forecasting, we often need to predict multiple future time steps. In a supervised learning context, this requires a strategy called iterative forecasting or rolling forecasting.
What is iterative forecasting?
In iterative forecasting, after predicting the next time step, you use that prediction as an input to forecast the following time step. This process is repeated until you have forecasted all the required future values. This has the benefit of using a single model to generate multiple forecasted values rather than a separate model for each future data point. This method poses two key challenges. The first challenge is a dependency on lagged features. As you generate future predictions, you must update the lagged features with the predicted values. The second challenge is error...
